This paper presents an extension to David Lowe’s well-known object recognition algorithm based on his Scale and Feature Invariant Transform (SIFT). One of the benefits of Lowe’s SIFT-based method is that it can recognize objects from only three keypoints. While this capability can be useful in circumstances where the cost of a false negative is high, it’s often the case that false positive are an equal, or greater, concern. We extend Lowe’s algorithm by adding the ability to use 3D constraints during matching. These constraints essentially eliminate false positive matches. We combine our extension with the original algorithm to retain the recognition power of the original method while adding significant robustness against false positives, thereby increasing overall classification power. In addition to improving recognition, our extension returns 3D pose information. Yet, it adds very little computational overhead to Lowe’s original algorithm.

Recent advances in computer vision have given rise to a robust and invariant visual pattern recognition technology that is based on extracting a set of characteristic features from an image. Such features are obtained with the scale invariant feature transform (SIFT) which represents the variations in brightness of the image around the point of interest. Recognition performed with these features has been shown to be quite robust in realistic settings. This paper describes the application of this particular visual pattern recognition (ViPR) technology to a variety of robotics applications: object recognition, navigation, manipulation, and human-machine interaction. The paper also describes the technology in more detail and presents a business case for visual pattern recognition in the field of robotics and automation

The first IEEE-IFR joint forum on Innovation andEntrepreneurship in Robotics and Automation tookplace on 10 April 2005 in Barcelona, Spain. The fol-lowing article was presented at the forum as one ofthe three nominees selected for the Innovation andEntrepreneurship in Robotics and AutomationAward. For further information, please visithttp://teamster.usc.edu/∼iera05/index.htm. Hadi Moradi

This paper presents a system for automatic reconbetween the left and right images, resulting in large errors struction of 3D structure using two unsynchronized cameras. in the final estimated 3D structure. This paper presents a Three images are acquired sequentially from the left, right, solution for correcting this change of pose in order to obtain and again from the left camera. A virtual image from the left camera synchronized with the right image is created by an accurate 3D structure. It also describes a fully functional interpolating matching points of interest (SIFT features) in the stereo vision system that typically finds corresponding points two left images. Both geometric and probabilistic criteria are in the two views with less than 1% outliers. The webcams used to select the correct set of matching features amongst the used in our system had additional wide-angle lenses attached three views. In an indoor environment, the method typically that presented a noticeable degree of distortion that was results in 3D structure with approximately 200 feature points, with a median 3D accuracy of 1.6 cm when the average depth removed using a well-known calibration toolbox 11], 12], 17]. is 3 m and the robot has moved 1-2 cm between each image Different approaches have been presented to synchronize acquisition. unsynchronized multi-view video streams, where the main

Performance of an online auction can be improved by an algorithm that learns from the set of bids already made. An online auction receives bids, and deals with each individually, deciding whether to accept a bid or wait for a higher one. This paper sets bounds on the performance of a learning algorithm, and shows that, for moderately long auctions, the performance is better than that of an existing algorithm. The proofs use formal mathematics, which are difficult to summarize. The flavor of the paper may be given by a simplification of one result: For any function f (h) o(hlog logh), where h is the ratio between the highest and lowest bids, if the sequences of bids made is greater than or equal to f (h), then the revenue generated is comparable to that of an optimal offline auction. The paper seems to be intended for those mathematically inclined, and does not focus on the mechanics of the algorithm.

Service robotics products are becoming a reality. This paper describes three core technologies that enable the next generation of service robots. They are low-cost, make use of low-cost hardware, and prepare for a short time-to-market for product development. The first technology is an object recognition system, which can be used by the robot to interact with the environment The second technology is a vision-based navigation system (vSLAM/spl trade/), which simultaneously can build a map and localize the robot in the map. Finally, the third technology is a flexible and rich software platform (ERSP/spl trade/) that assists developers in rapid design and prototyping of robotics applications.

The problem of detecting and labeling a moving human body viewed monocularly in a cluttered scene is considered. The task is to decide whether or not one or more people are in the scene (detection), to count them, and to label their visible body parts (labeling). It is assumed that a motion-tracking front end is supplied: a number of moving features, some belonging to the body and some to the background are tracked for two frames and their position and velocity is supplied (Johansson display). It is not guaranteed that all the body parts are visible, nor that the only motion present is the one of the body. The algorithm is based on our previous work [12]; we learn a probabilistic model of the position and motion of body features, and calculate maximum-likelihood labels efficiently using dynamic programming on a triangulated approximation of the probabilistic model. We extend those results by allowing an arbitrary number of body parts to be undetected (e.g. because of occlusion) and by allowing an arbitrary number of noise features to be present. We train and test on walking and dancing sequences for a total of approximately 104 frames. The algorithm is demonstrated to be accurate and efficient.

Decomposable triangulated graphs have been shown to be efficient and effective for modeling the probabilistic spatio-temporal structure of brief stretches of human motion. In previous work such model structure was handcrafted by expert human observers and labeled data were needed for parameter learning. We present a method to build automatically the structure of the decomposable triangulated graph from unlabeled data. It is based on maximum-likelihood. Taking the labeling of the data as hidden variables, a variant of the EM algorithm can be applied. A greedy algorithm is developed to search for the optimal structure of the decomposable model based on the (conditional) differential entropy of variables. Our algorithm is demonstrated by learning models of human motion completely automatically from unlabeled real image sequences with clutter and occlusion. Experiments on both motion captured data and grayscale image sequences show that the resulting models perform better than the hand-constructed models.

In the future, people will interact with computers in a vastly different way than they do now. Computers will be able to see and hear people, interpreting their motion, moods and speech. Computers will also communicate back to us through embodied virtual characters that can themselves move, gesture, and speak. This thesis deals with two important components of such a human-machine interface technology: the automatic observation and synthesis of human motion. In the first part of this thesis, a real-time system for visually tracking the 3-D motion of an arm without any special markers or sensors attached is described. The system can be used as a human-machine interface and can estimate three DOFs for the shoulder and one DOF for the elbow from a monocular view taking into account perspective projection. The robustness and efficiency of the system is enhanced by avoiding explicit feature extraction, and instead encoding image properties directly as a function of the degrees of freedom of the arm model that are to be estimated. The system runs at 30 Hz and tracks natural arm motion with 1 cm of resolution and accuracy in 3-D. In the second part of this thesis, a system for automatic synthesis of realistic human motion is described. The techniques developed are computationally efficient and can be implemented to synthesize motion in real-time at interactive rates. The main principle of the method is to describe human motion in terms of actions parameterized by state, goal, and style parameters. Mathematical models are learned using machine learning techniques based on observations of real people performing actions. The resulting models represent the motions accurately because of the careful choice of the space in which to represent motion; rather than work in the joint angle space where position and orientation errors accumulate as one progresses down the skeletal hierarchy, motion is represented in the 3-D Euclidean space of the motion-capture markers used to acquire the sample motions. In real-time it is then possible to map the synthesized motion to corresponding (error minimizing) joint angles with the use of a Kalman Filter.

Computer perception of biological motion is key to developing convenient and powerful human-computer interfaces. Successful body tracking algorithms have been developed; however initialization is done by hand. We propose a method for detecting a moving human body and for labeling its parts automatically. It is based on maximizing the joint probability density function (PDF) of the position and velocity of the body parts. The PDF is estimated from training data. Dynamic programming is used for calculating efficiently the best global labeling on an approximation of the PDF. The computational cost is on the order of N 4 where N is the number of features detected. We explore the performance of our method with experiments carried on a variety of periodic and non-periodic body motions viewed monocularly for a total of approximately 30,000 frames. Point-markers were strapped to the joints of the subject for facilitating image analysis. We find an average of 2.3% labeling error; the experiments also suggest a high degree of viewpoint-invariance

We have developed a system capable of tracking a human arm in 3D and in real time. The system is based on a previously developed algorithm for 3D tracking which requires only a monocular view and no special markers on the body. In this paper we describe our real-time system and the insights gained from real-time experimentation